Civilization's Blind Spot (kunnas.com)

36 comments

route_missing5 comments

The strongest sentence is “what can be seen without becoming anybody’s work.”

In large organizations we tend to call these communication failures, but often communication has succeeded perfectly. Everyone relevant has seen the problem. Several people agree it is real. What is absent is an owner with the mandate, budget, and political cover to do anything about it.

The engineering analogy would be an issue tracker. An observation becomes organizationally real when it has a stable identifier, a component, an assignee, and eventually a disposition. Public policy has lots of specialized trackers—courts, regulatory proceedings, grant calls—but no global repository for “this seems important and does not fit an existing component.”

The analogy breaks at merge rights, though. Filing a ticket does not create legitimate authority to resolve it. The diagnosis is persuasive; the interesting part is whether the missing thing is routing, authority, or both.

ptr_to_void2 comments

I don’t read the proposal as “create an inbox for good ideas.”

The useful distinction is between receiving a message and accepting responsibility for its disposition. Most citizen-input systems do the first. Acquisition, in the author’s terminology, means somebody must classify it, refer it to a recipient who accepts the referral, or close it with a reason.

In distributed-systems terms, this is closer to acknowledged delivery than a suggestion box. Governments have many endpoints, but most accept only packets matching a schema they already recognize.

underlapcollapsed

Acknowledged delivery presupposes an address space.

The difficult cross-boundary claims are frequently saying that the current address space is wrong: the problem is produced by the division between health, housing, policing, and social services, for example. A better router cannot invent a service whose jurisdiction is “the interaction among four existing services.”

At some point the router itself becomes the institution with general jurisdiction, which recreates the capture and overload problem one level up.

old_man_yells_at_queue2 comments

Every sufficiently large company eventually builds an “innovation portal.” Employees submit ideas, receive an automated acknowledgment, and the ideas disappear into a database rather than an inbox.

The scarce resource was never identifiers. It was attention from someone able to reallocate resources. If you impose a duty to give reasons, you will get a dropdown containing twelve standardized reasons.

Bureaucracies are very good at transforming substantive obligations into form completion.

route_missingcollapsed

That failure mode seems real, but even standardized rejection creates data that silence does not.

You could discover that 70% of cross-department submissions are being closed as “outside mandate,” that one office rejects almost everything, or that the same proposal is independently rediscovered every two years.

It would not make the institution wise. It would make one particular kind of failure inspectable. That is a smaller claim than the essay’s title, but still useful.

plaintextonly5 comments

I almost stopped at the “corpus frame.”

Every noun has a capitalized companion essay. We have mechanism-space, reference telos, adaptive safety margin, metamechanisms, and a declaration that democracy and rights are mechanisms rather than terminal values, all before reaching the article.

Once it gets to Dunant, the writing becomes concrete and quite good. The actual thesis could probably be communicated in one third the length: reforms need a chain of people and institutions; history remembers the chains that completed; things that never enter an institutional process are hard to count.

This feels like a useful paper trapped inside the documentation for a philosophy framework.

eigenvalue2 comments

I had the same initial reaction, but the excessive framing may be an adaptation to internet criticism.

The author explicitly pre-answers “what about power,” “what about survivorship bias,” “aren’t most unsolicited proposals bad,” and “aren’t you doing Great Man history.” Without that, the first hundred comments would consist of people raising those objections as though the author had not considered them.

It makes the page less readable, but also unusually difficult to knock over with the standard one-paragraph response.

soft_constraintscollapsed

It does not neutralize those objections. It declares a philosophical answer to them.

Saying rights and democracy are “mechanisms under evaluation” rather than terminal values is a very strong consequentialist commitment. It is not methodological housekeeping. A system that optimizes a civilizational safety margin may quite rationally decide that my rights are an expensive local optimum.

The narrow historical argument does not require any of that machinery, which is another reason the preamble distracts.

power_is_a_feature2 comments

My larger objection is that the article treats power as something that blocks a correction after the correction has become legible.

Classification is already an exercise of power. Calling something a correction, grievance, conspiracy theory, private dispute, technical defect, or political demand determines who must engage with it and what evidence is admissible.

There is no neutral intake layer before politics. The intake layer is where a great deal of politics happens.

supplychainofideascollapsed

The article defines “correction” as an input rather than a certified truth, so I think it partly sees this.

But I agree that a real implementation would need to preserve the history of classification itself: who labeled the claim, what alternative labels were considered, which vocabulary was required, and which populations repeatedly failed to produce “well-formed” submissions.

Otherwise the new system merely converts social illegibility into procedural illegibility.

sample_of_successes4 comments

The historical examples establish less than the author thinks.

Dunant, Nightingale, Lemkin, Nader, and the others succeeded through networks containing investigators, advocates, political sponsors, implementers, and institutions. Yes. But every nontrivial accomplishment has a chain of complementary roles. A successful restaurant also requires suppliers, cooks, landlords, inspectors, and customers.

Showing that reform required a chain does not show that civilization is missing a standing chain, or that a standing chain would have selected these proposals before a crisis did. It might instead formalize the prevailing assumptions and reject the unusual proposal more efficiently.

The article is admirably explicit that the cases are purposively selected and cannot estimate lost wisdom. But that concession leaves a large gap between “biography hides coordination” and “civilization has a blind spot requiring new architecture.”

institutional_memory2 comments

I think the nontrivial contribution is changing the unit of analysis.

The usual story asks why Dunant succeeded: moral insight, persistence, timing, charisma. The essay asks which transitions had to occur and which role owned each transition. That produces a different question: why did one person have to locate and connect all those roles manually?

It does not prove that a general intake institution would work. It does show that “the right idea won” is an inadequate causal account.

sample_of_successescollapsed

Agreed. “Biography is lossy compression of institutional labor” is the strongest part of the piece.

I would have preferred the essay to stop at that historiographic finding, then present the general-intake idea as a hypothesis to test. The current version moves between diagnosis and design with more confidence than the evidence supports.

counterfactualistcollapsed

There are narrower domains where this could be tested.

Aviation near-miss reporting, pharmacovigilance, vulnerability coordination, and industrial incident databases all created standing paths from weak signals to owned cases. One could examine what kinds of problems became visible before and after those systems existed, how much noise they generated, and which organizational safeguards mattered.

“Civilization” may be too large a unit for evidence. Start with domains where intake architecture changed while the underlying people remained roughly comparable.

civic_stack5 comments

I worked on municipal systems and saw a very literal version of this.

We found an accessibility problem caused by the interaction of procurement templates, the legal review process, an identity platform, and individual service departments. Everyone agreed the result was bad. Procurement said it did not own user experience. IT said it implemented requirements. Legal said it reviewed compliance, not service design. The departments said the shared platform was outside their control.

Nine months were spent finding someone senior enough to convene four groups and keep the issue alive. The technical changes took about two weeks.

The person who finally carried it was exactly a “temporary institution”: case manager, translator, project manager, political sponsor, and memory system. Once she changed jobs, similar issues began falling through the same gap again.

mandarins2 comments

From the civil-service side, central coordinating offices and cross-government task forces exist precisely for this reason.

The problem is not usually that nobody can be identified as an owner. It is that ownership carries risk and rarely carries reward. A cross-cutting problem is an opportunity to inherit accountability for outcomes you do not control.

“Nobody owns it” often means “nobody powerful benefits from being recorded as its owner.”

civic_stackcollapsed

That seems compatible with the article.

A crisis or senior sponsor changes the payoff and suddenly all the required roles can be occupied. Before that, the issue is technically visible but institutionally inert.

The design problem may be less “find the correct box on the org chart” and more “make accepting a legitimate referral safer than ignoring it.”

orgchartmaximalistcollapsed

This is also what senior individual contributors do in very large technology companies.

The glamorous description is “provides technical leadership across organizational boundaries.” The less glamorous description is “manually supplies the interfaces that the organization omitted.”

We celebrate these people because admitting the company repeatedly requires heroics would suggest the architecture is defective.

standardswonkcollapsed

Standards organizations are a partial counterexample.

An individual can submit a draft, it gets a stable name, discussion is archived, objections are recorded, and the proposal can move through defined states without the author first obtaining executive sponsorship.

But that works because the domain is bounded, the community shares a technical vocabulary, and adoption is often voluntary. A proposal addressed to “society” has no comparable scope boundary or interoperable definition of correctness.

The more general the intake mechanism becomes, the less obvious it is what counts as a valid packet.

bayes_lawyer5 comments

The “missing denominator” argument is clever and also perilously close to unfalsifiable.

If valuable proposals disappear before becoming institutional cases, then by definition we cannot observe how many there were. But what empirical result would distinguish:

1. civilization discards a large amount of important knowledge, from
2. civilization mostly ignores redundant, self-interested, or incoherent submissions?

The author says the claim is failed calibration rather than a numerical estimate. Fine, but lack of calibration is not evidence that the current filter performs badly. A spam filter does not produce a reasoned judgment for every deleted message and may nevertheless be excellent.

missing_not_random2 comments

It is evidence that you cannot estimate the error rate from the retained sample.

That does not imply the error rate is high. It means success stories cannot establish that it is low.

In statistical terms, the archive is selected by a process correlated with the outcome being studied. The essay’s historical claim is mostly negative: do not infer that civilization reliably recognizes good reforms from a dataset consisting of reforms it recognized.

bayes_lawyercollapsed

I accept that as a warning about historical inference.

I am less convinced it supports an institutional duty to retain and adjudicate more claims. Measurement has a cost. A perfectly observable filter that spends the entire budget documenting garbage can be worse than an opaque filter that leaves capacity for implementation.

trialballooncollapsed

This can be made prospective rather than historical.

Take a random sample of submissions that existing processes would discard or leave unanswered. Give them independent routing and limited review. Track duplicates, subsequent rediscovery, adoption, measurable improvements, and reviewer disagreement.

You would not learn the value of every invisible proposal, but you could estimate whether the neglected stream contains enough signal to justify additional capacity.

The initial experiment could be tiny. The system does not have to promise universal adjudication in order to measure its blind spot.

archivist42collapsed

Complete disappearance may also be rarer than the essay implies.

Private papers, organizational archives, meeting notes, local newspapers, and oral histories often reveal that an allegedly novel reform had predecessors. Historians routinely find warnings that were recorded somewhere but never connected to a decision process.

That still supports the institutional-acquisition argument, but it changes “there is no object” into “the objects are fragmented and expensive to reconstruct.” The latter is less dramatic and possibly more tractable.

rate_limit_everything5 comments

A general correction endpoint would be the largest denial-of-service vulnerability in government.

Today it would receive sovereign-citizen filings, perpetual-motion machines, miracle medical protocols, personalized tax theories, and allegations that a neighbor’s router is controlling the weather.

With language models, each person can submit 10,000 polished, citation-shaped policy briefs instead of one handwritten manifesto. A duty to classify or answer them turns inexpensive text generation into expensive public-sector labor.

The article correctly observes that AI may enlarge the queue faster than review capacity. I think that is not one concern among many. It is the central technical constraint.

typed_channels2 comments

We know some ways to make intake abuse-resistant:

typed submissions, duplicate detection, refundable deposits, rate limits, submitter reputation, public queues, randomly audited rejections, and progressively higher evidentiary requirements as a claim advances.

Courts do not respond to every email, but they do provide a route through which a properly formed claim can become a case. The interesting design question is what the minimal form should be for a cross-boundary public claim.

public_choicecollapsed

“Properly formed” is where the institution captures itself.

Over time, professionals define novelty, acceptable evidence, standing, and relevance in ways that make submissions from other professionals easy to recognize. The intake mechanism becomes a guild with a form attached.

Deposits also select for wealth, reputation systems select for incumbency, and typed channels exclude the claims whose point is that the existing taxonomy is wrong.

llm_burner2 comments

The asymmetry is worse than ordinary spam.

Producing a plausible eighty-page cross-domain proposal may cost cents. Determining that its citations are real, its causal model is coherent, and its implementation does not contradict five bodies of law may require expert-days.

You cannot solve that with summarization because the expensive part is verification. A model can make the queue easier to read while making it much larger and more adversarial.

queueingtheorycollapsed

That suggests sampling rather than universal service.

Keep the normal high-precision filters, but draw a statistically meaningful sample from the rejected stream for expensive review. Publish the categories in which reviewers repeatedly discover false negatives.

You get calibration and pressure to improve the filter without promising every submitter a bespoke administrative proceeding.

price_signal2 comments

Isn’t the standing path from neglected observation to coordinated action usually called a market?

Someone sees a problem, assembles the missing capabilities, bears the risk, and captures part of the value if the solution works. The entrepreneur is a temporary institution until the company becomes a permanent one.

The examples here are mostly public goods or political changes where the value cannot be appropriated. That is important, but it makes the scope considerably narrower than “civilization’s blind spot.” It is the familiar problem that diffuse benefits and concentrated costs do not generate a buyer.

externality_budgetcollapsed

“Mostly public goods and political changes” still covers law, safety, standards, environmental externalities, institutional design, and long-horizon risks.

The same issue also appears inside firms. An improvement may create substantial company-wide value while offering no department enough local return to fund it. Internal markets do not help when accounting boundaries are what make the proposal ownerless.

Markets route corrections when value is excludable and somebody has authority to purchase the change. The essay is largely about cases where those preconditions are absent.

kingdonswindow2 comments

This reads like Kingdon’s multiple-streams framework, agenda-setting research, and organizational absorptive capacity translated into systems language.

There is a problem stream, a policy stream, and a political stream. A policy entrepreneur couples them when a window opens. Organizations recognize and assimilate external knowledge only when they have relevant prior capacity. Demands can also be kept outside the agenda entirely.

The article cites much of this, so it is not pretending the components are new. But I wanted a more explicit account of the delta: which proposition is established by the existing literature, which is the author’s synthesis, and which is a new empirical claim?

bibliophagecollapsed

I think the added claim is about the boundary before those literatures obtain a stable object.

Agenda-setting research can study an excluded demand when researchers can identify the demand and the arena that excluded it. The essay points to claims that never become demands addressed to any arena, so even “nondecision” may be too advanced a state.

Whether that is a genuinely new category or a relabeling of archival silence is debatable. A comparison table would have helped more than another layer of corpus terminology.

constitutional_bug3 comments

The diagnosis persuaded me more than the linked solution.

The proposed Mechanism Authority is supposed to evaluate incentives, long-run effects, metric distortion, institutional drift, and essentially anything affecting civilizational persistence. It cannot directly veto policy, but its findings can force disclosure, delay, and a formal political response.

That seems like recreating the problem in one enormously consequential intake institution. Its blind spots become constitutional blind spots. Its choice of what to examine becomes the new scarce resource. And because its mandate is “does this mechanism actually work?”, almost every political disagreement can be restated as a technical disagreement within its jurisdiction.

I would start with a much smaller public router: stable case identity, jurisdiction search, accepted referral, reasoned closure, and random audit. Do not centralize substantive truth-finding unless narrower versions demonstrate value.

separation_of_concerns2 comments

The absence of a veto matters.

Requiring Parliament to state why it is proceeding despite an adverse analysis is closer to a budget office, ombudsman, or environmental review than rule by experts. Elected officials retain the decision; the institution raises the cost of pretending a warning did not exist.

We already tolerate independent bodies that structure the information and timing around political choices.

constitutional_bugcollapsed

Information and timing are forms of power.

That does not make the institution illegitimate, but “it only forces an answer” understates the stakes. Parliamentary attention is finite. Being able to place an item on the compulsory-answer queue means deciding which warnings become politically expensive to ignore.

The essay’s best insight applies recursively: who receives corrections about the correction authority?